YOLO26n 堆高機偵測 ｜ cvat #9 ｜ 訓練/推論 imgsz 1280 ｜ 並列觀察
unit=boxes · 總 16,765 框(train 13,006/val 1,541/test 2,218)· 7 廠商
| 廠商 | 鏡頭 | 總框 | train | val | test |
|---|---|---|---|---|---|
| PUBLIC | 7 | 9,656 | 7,301 | 622 | 1,733 |
| IRODA | 3 | 2,531 | 1,910 | 417 | 204 |
| FOX | 1 | 1,661 | 1,353 | 205 | 103 |
| HONCHUAN | 4 | 1,528 | 1,211 | 139 | 178 |
| unknown | 1 | 1,221 | 1,131 | 90 | 0 |
| IRODA-AI | 2 | 120 | 68 | 52 | 0 |
| FOX-AI | 1 | 48 | 32 | 16 | 0 |
訓練資料:實吃 14,919 → 16,765(+1,846 框)
| 指標(同test) | v702 | v717 | Δ |
|---|---|---|---|
| mAP50-95 | 0.843 | 0.8489 | +0.0059 |
| mAP50 | 0.9873 | 0.994 | +0.0067 |
verdict:全面勝v702,已上版並列